Skip to main content Skip to navigation

Controlling for Fixed Country Characteristics

A final refinement in the construction of our index is the following. Broadly speaking, the variables used in our index to measure different dimensions of globalisation measure outcomes, rather than policy. For example, the variable they use to measure openness to trade is the value of total trade (imports plus exports) as a percentage of GDP.

A well-known problem with this measure of trade openness for a given country is that it depends not only on underlying trade policy (i.e. tariff and non-tariff barriers to trade imposed by the country in question) , but also on the geographical and economic characteristics of a country. Other things equal, countries with large populations and diversified economies will trade less (as a proportion of GNP) than small countries. For example, both the Netherlands and the US are highly open to trade in the sense that they have low tariffs and non-tariff barriers. However, in 2001, the trade openness scores for the Netherlands and the US were 129% and 23% respectively. But is the Netherlands really over five times more open to trade flows than the US?

We feel that this problem is most serious for the variables that make up the economic globalisation index, as it is here that variation in country size (population or land area) or geographical location is most likely to affect the economic outcome, given a fixed policy stance.

There are then two possible solutions to this problem. First, rather than measuring outcomes, we could  try to measure the underlying policies directly. However, except in the case of trade openness[1], the data are not available to do this. For example, there are no widely available[2] quantitative measures of the strength of “capital controls” i.e. controls on the capital account that can affect inward or outward direct or portfolio investment.

A second solution is to correct the outcome measure of openness (i.e. trade flows, FDI and portfolio investment flows, income payments and receipts)  for relevant country characteristics. This is done[3] by least-squares regression of any one of the openness measures on a number of country characteristics that are thought to be: (a) exogenous to economic openness; and (b) relevant in determining economic openness as a percentage of GNP.

The resulting residuals (actual value minus the predicted value) then measure the extent to which a country is more or less open than would be expected, given its characteristics. So, we interpret the residuals from the regression as the “corrected” or “adjusted” measure of openness.

Here, we apply this method of  adjustment to all the measures of economic openness (i.e. trade flows, FDI and  portfolio investment flows, income payments and receipts).   That is, our measures of these variables are the residuals from the regressions just described.

We now describe the regressions in a bit more detail. Our choice of relevant country characteristics were: population in 1998 (POP), land area (AREA), and a dummy variables recording whether the country was landlocked (LANDLOCK). This dummy is included as countries without seaports face higher costs of international trade, and this may well affect foreign direct investment. Indeed, Sachs(2001) finds that distance from the sea-coast is negatively related to per capita GDP. Finally, we do not include the usual measure of economic development, GDP per capita, although its inclusion would undoubtedly increase the explanatory power of our regressions.

The reason is the following. In our view, what our index is ultimately trying to measure is to what extent the past and present policy choices of a country have led it to integrate with the world economy (and society). These policy choices (given geographical characteristics) also determine its level of economic development (as measured by GDP per capita). So, “stripping out” the effects of GDP per capita from the various measures of globalisation would in fact be removing valuable information from these measures. The regression results are described below.

 

Trade
FDI
Portfolio
Income

Log AREA

-1.57***

(0.47)

-0.197***

(0.05)

-0.21**

(0.10)

-0.01

(0.14)

Log POP

-14.43***

(0.50)

-0.23***

(0.061)

0.04

(0.11)

-1.47***

(0.15)

LANDLOCK

-11.06***

(2.22)

-0.56**

(0.26)

-0.55

(0.54)

0.16

(0.67)

Adjusted R-sqr

0.272

0.015

0.001

0.035

Robust standard errors in brackets: *** = significant at 1%, **= significant at 5%.

In the trade and FDI regressions, AREA, POP, and LANDLOCK are individually significant, and in some cases have quite large effects.  Jointly, these three factors explain about 27% of the total variation in the trade variable, although the percentage of the total variation in FDI explained  is much lower, reflecting the fact that FDI flows are highly volatile

For example, if a country is landlocked, its trade variable is 11 percentage points lower than it would be otherwise. Recalling from Table 1 that the mean value of the trade variable is about 83%, this means that other things constant, a landlocked country has 13%. less trade with the rest of the world than it would otherwise have. Again recalling from Table 1 that the mean value of the trade variable is about 2.35%, a similar calculation (divide 0.56 by 2.35) implies that a landlocked country has about  24%  less FDI flows with the rest of the world than it would otherwise have.


[1] Data is available on average tariffs and non-tariff barriers  from the World Bank.

[2] Such measures have been constructed by various authors for a limited set of countries and years using qualitative information in the IMF’s publication, Exchange Controls and Exchange Restrictions: see e.g.  (Quinn(1997)).

[3] This approach extends Pritchett(1996), who applied it to get various measures of trade openness .